Multimodal memory and perception. Real-time vision super-resolution. Beyond attention for long contexts. Do LLMs truly reason?. Rails for accountable agents